# Harbor Example Research — fictional house documentation policy **Entire house layer: simulated scenario input.** Harbor Example Research is a fictional institution created for this demonstration. These conventions apply only to the commissioned document. They are not rules from a regulator or the policies of an actual financial institution. ## Document front matter Title: `USD SOFR OIS Discount Curve — Methodology Document`. Include document version `demo-v1`, actual document date, as-of date `2026-09-18`, status `for documentation readiness assessment`, and fictional inventory ID `HER-QM-017` with `simulated` adjacent to it. Copy this disclaimer verbatim: > This methodology demonstration combines real historical public data, derived numerical evidence and an explicitly fictional institution dossier. It is not a live pricing service, an approved production model or a complete regulatory submission. All institutional roles, policies, records and controls are simulated; actual institutional evidence and authorization remain outside this demonstration. Give a three-sentence plain-language conclusion: what the model does, why the developer chose the principal method, and its material tradeoff. A rounded money figure may appear here if its single-cash-flow basis is explicit. Add a concise model overview and a short reading guide to the principal comparison, numerical evidence and limitations. ## Twelve-section template Use these exact `##` headings, with `###` subheadings as useful: 1. Purpose and scope 2. Product and portfolio coverage 3. Model description and theory 4. Inputs and data 5. Method choices and rationale 6. Implementation notes 7. Testing and benchmarking 8. Limitations and compensating controls 9. Assumptions register 10. Governance and institutional record 11. Third-party components 12. References The document describes the model and the selected fictional scenario. It does not narrate its drafting process. Keep the section mapping, claim audit and readiness checklist in `reviews.md`. Cover the whole scope substantively; twelve empty headings or a compact decision note are insufficient. ## Evidence presentation Tables have a caption, units in each numerical column, the as-of date, a population/grid definition where relevant, and a source line naming files and calculation scripts. Identify percent, basis points, currency and dimensionless discount factors separately. One basis point is 0.0001 in rate units. Do not turn a rate difference into a cash amount without a specified cash-flow measure. Figures have axis labels with units, a caption stating the historical as-of date, calculation script and frozen data source. Use at least the supplied discount-factor or zero-rate comparison, forward comparison, and either the 5Y bump or cash-flow sensitivity ladder where they support discussion. Explain what the plot demonstrates and what it does not. Copy used figures locally into the output directory; do not label the old plots as newly generated. Citations use short keys plus specific file, section, table, JSON field or CSV row/column locators. Expand every key in References with a public URL. Use public source URLs for substantive background, and the developer brief for supplied reasoning. References need only list sources actually used. Do not quote long passages from third-party papers; paraphrase with precise attribution. ## Modeling choices and tests The principal method-choice discussion starts by defining the largest observed present-value disagreement between A and B for USD 100 million paid on one monthly-grid date. Present a comparison table with methods A and B, criteria, measures and source locators. Discuss methods measured only on a subset of criteria in their own paragraph or table. State the preferred method as the developer’s choice, not as independently established truth. The fictional house repricing criterion is absolute par-rate error at most 0.01 basis point for **every** calibration instrument. Include or link a table with all 22 instruments and both methods’ errors; reporting only the maximum is insufficient. The held-out median absolute residual criterion is at most 1 basis point for the selected population. Report the actual historical values against these supplied criteria without interpreting them as production acceptance. For method A’s separately written implementation comparison, give maximum and mean absolute discount-factor differences and the exact retained business-day grid: 7,482 observations from 2026-09-18 through 2056-09-26 inclusive. Do not rename it a calendar-daily or monthly grid. Distinguish this implementation check from A-versus-B method disagreement on a monthly grid. Include the descriptive end-date distance bins with medians and interquartile ranges for both held-out cohorts; distance is not a measure of independent validation or total financial risk. Explain the historical paper-example and round-trip checks, repricing, convention sensitivities, continuity, positivity-collar behavior, extrapolation and bump locality. Distinguish empirical properties on this snapshot from general guarantees. Identify failed variants as failed where reported; do not omit the QuantLib paper-setting solver failure. ## Institutional facts, language and limits Use `[INSTITUTIONAL GAP: precise missing evidence]` on its own line for every selected actual institutional fact that the fictional dossier cannot establish. The dossier supplies a scenario inventory, ownership, proposed policy and controls; it does not fill actual approval, operating effectiveness, real benchmark or deployment evidence. Do not turn a fictional named role into an actual reviewer. Prefer `discount curve`, `node`, `instrument`, `par rate`, `as-of date`, `readiness check`, `separately written implementation`, and `pricing outside the input set`. Do not call the developer’s own comparison an independent institutional validation. State dates in ISO form, preserve supplied numerical precision, and use eight decimal places for displayed discount factors when supported. Explain acronyms at first use. Keep paragraphs focused. Consolidate repeated figures in their owning section and cross-reference elsewhere. Preserve exact population, interval and sign conventions even when rounding. No statement that a test passed today is allowed unless the current executor actually performed that test within the permitted documentation scope. ## Required data notices Retain the complete SOFR data republication notice and nonaffiliation statement from [the source register](https://openprose.ai/demos/sofr/v1/sources.md#sofr-data-republication-notice) in the document’s References section, with the linked New York Fed terms. This provider notice accompanies the historical reference-rate data and is distinct from the simulated institutional disclaimer.